woodson

For a long time, AI was a bad word that stood for unfulfilled promise. See AI Winter. Hence, researchers strictly avoided the term while still working on learning algorithms, the same that power LLM training.

There’s a bit of a grey area, for example speech recognition. Would you classify that as analytic or generative? Whisper and speech LLMs work pretty well, but can completely make up stuff that wasn’t in the audio at all (see e.g. “thank you for watching” transcribed during silence). Other approaches are closer to the acoustic evidence but may make other mistakes (especially wrongly transcribing long tail, low frequency terms). Pick your poison.

Thanks for sharing, this is a great improvement (also and especially for human use! Renders much better on mobile phones compared to Confluence).

Alongside the poorly lit spaceship. Spaceships are workplaces and workplaces should provide adequate illumination so you can see what you are doing.

I always thought the same about CSI. Have the set designers ever been to a lab?

5090 ($2k MSRP but realistically $3-3.5k)

These days, more like >$4.1K (at least in the US).

Yeah, there have been several papers with attacks on Whisper:

- Inject adversarial noise to make it transcribe what you want (https://arxiv.org/abs/2210.17316)

- Stop it from transcribing (https://arxiv.org/abs/2405.06134)

- Adversarial prompt injection to make it translate instead of transcribe (https://arxiv.org/abs/2407.04482v2).

Indeed. And the one in Admont, Austria: https://stiftadmont.at/en/about-the-abbey-library/

HF is notorious for making it difficult to work offline (or at least not waste time trying to connect when everything needed is offline) and is constantly changing how it is being handled. Previously, there was TRANSFORMERS_OFFLINE, HF_DATASETS_OFFLINE, etc.

You can use devcontainers (in VSCode or separate), like this: https://github.com/entn-at/claude-rust-devcontainer/

This will limit the agent in what it can do in the system and what IPs/domains it can reach. This requires a lot of customization to your specific framework/environment. Note that this can reduce the agent’s effectiveness, as it will have to “work around” some of the limitations. This isn’t foolproof either, and the agent could exfiltrate data e.g. via DNS requests.

But I don't see a scenario where pytorch needs network access.

Training models across multiple compute nodes? That’s a big one.

Not if one is running it in a non-privileged vm/container with restricted network access. But everything is YOLO these days.

Why generative? Or has it been decided that only generative models are “AI”?

It highly depends on the sort of data you’re processing (phone calls, podcasts, meetings of more people recorded using single channel?). For NVIDIA/NeMo, check out their softformer diarization models (also streaming).

Well, but their customers are those that buy Apple hardware.

Look into RWKV.

It’s a distinction that IMHO likely doesn’t make much difference, at least for the mostly automated/non-interactive coding agent use case. What matters more is how well the post-training on synthetic harness traces works.

There are still opportunities, but they aren’t paid nearly as well as less researchy positions in industry. US post-doc salaries at state universities aren’t that high.

You mean Moshi (https://github.com/kyutai-labs/moshi)? Since Personaplex is just a finetuned Moshi model.

…who lives in Oregon, in the US.

This is about speech to text, not speech recognition.

You still need a tax accountant in France to register the FTE and file paperwork with the tax office and social insurance.

The bank issues/refusals may have something to do with FATCA. If you have anything to do with the US in terms of taxes, many EU banks don’t want you as their customer. If it’s a subsidiary of a foreign company, then a lot of paperwork is required to prove that the foreign owners actually exist.

This is basically a devcontainer, right?

Jeez, OP asked what it means in this context (zero-shot voice cloning), where you quoted a generic definition copied from Wikipedia. I defined it concretely for this context. Don't take it as a slight, there is no need to get all argumentative.

From your definition:

a learner observes samples from classes which were not observed during training, and needs to predict the class that they belong to.

That's not what happens in zero-shot voice cloning, which is why I dismissed your definition copied from Wikipedia.

I don't disagree, but that's what people started calling it. Zero-shot doesn't make sense anyway, as how would the model know what voice it should sound like (unless it's a celebrity voice or similar included in the training data where it's enough to specify a name).

Does the 169M include the ~90M params for the Mimi codec? Interesting approach using FiLM for speaker conditioning.

This generic answer from Wikipedia is not very helpful in this context. Zero-shot voice cloning in TTS usually means that data of the target speaker you want the generated speech to sound like does not need to be included in the training data used to train the TTS models. In other words, you can provide an audio sample of the target speaker together with the text to be spoken to generate the audio that sounds like it was spoken by that speaker.

The local Gemma models are pretty good for tasks involving multilingual inputs (translation, summarization, etc.). They have their niche.

There are perceptual hashing algorithms for images/video/audio (dsp and ML based) that could work for that.